Plug-and-Play Explanations for Diverse AI Models

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing methods struggle to provide universal explanations for various artificial intelligence models, especially in complex scenarios, and existing frameworks like TensorFlow and PyTorch only explain deep learning models within their specific platforms, lacking general applicability and accuracy.

Innovation Solution

A plug-and-play system comprising a service provider server with explanation modules and a plug-and-play manager that automatically identifies and connects an appropriate explanation module to a user's AI model, allowing customizable explanations through a user terminal.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Device complexity

If a single explanation method is used for all AI models, then the system complexity is reduced, but the explanation accuracy decreases because different AI models have different characteristics

Engineering Contradiction:
Improvesystem complexityVSAvoidexplanation accuracy
Core Design Contradiction:
Device complexityVSMeasurement precision

Solution Approach 1:

The explanation system is segmented into multiple specialized explanation modules, each designed to handle specific AI model types (decision trees, deep learning models, Bayesian models). This segmentation allows each module to provide accurate explanations tailored to its specific model type while maintaining overall system manageability through modular architecture.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent creates a universal explanation framework that can handle multiple AI model types through a common interface and selection mechanism. The system provides a unified way to request explanations while automatically selecting the appropriate specialized module, achieving both universality and specificity.

Inventive Principle:
Principle #6Universality (Multi-functionality)

2Measurement precision

If separate explanation algorithms are provided for each AI model type, then the explanation accuracy is improved, but the device complexity increases

Engineering Contradiction:
Improveexplanation accuracyVSAvoidsystem complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The explanation system is divided into independent, specialized modules for different AI model types. Each module can be developed, tested, and maintained separately, which improves accuracy for each model type while reducing overall system complexity through modular design and independent development.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

An intermediary selection mechanism acts as a mediator between the user's explanation request and the appropriate specialized explanation module. This intermediary component automatically identifies the model type and routes the request to the correct module, simplifying the user interface while maintaining specialized accuracy.

Inventive Principle:
Principle #24Intermediary (Mediator)

3Measurement precision

If existing deep learning frameworks explain only their own models, then the framework-specific explanation accuracy is improved, but the adaptability decreases

Engineering Contradiction:
Improveframework-specific explanation accuracyVSAvoidmodel compatibility
Core Design Contradiction:
Measurement precisionVSAdaptability or versatility

Solution Approach 1:

The patent creates a universal explanation platform that can interface with multiple deep learning frameworks (TensorFlow, PyTorch, etc.) through standardized interfaces. This allows the system to provide accurate explanations for models from different frameworks without requiring separate specialized systems for each framework.

Inventive Principle:
Principle #6Universality (Multi-functionality)

Solution Approach 2:

The system introduces intermediary layers that act as translators between different AI frameworks and the explanation modules. These intermediaries convert framework-specific model representations into a unified internal representation that can be processed by the explanation modules, enabling cross-framework compatibility while maintaining explanation accuracy.

Inventive Principle:
Principle #24Intermediary (Mediator)

Data Source

PatentEP4636653A1Plug-and-play-based method for providing description of artificial intelligence model
Publication Date: 2025.10.22 KOREA ADVANCED INST OF SCI & TECH
  • EP4636653A1 patent drawingFigure 1
  • EP4636653A1 patent drawingFigure 2
  • EP4636653A1 patent drawingFigure 3~4A

AI summary

A method of providing an explanation of an artificial intelligence model based on a plug-and-plug mode includes a service provider server setting a type of an artificial intelligence model, a plug-and-play manager recognizing an artificial intelligence model and creating a list of available explanation modules on the basis of a recognition result when the artificial intelligence model is input, the service provider server providing key information on explanation modules to a user terminal, the user terminal selecting one of the explanation modules and notifying the service provider server of the selected explanation module, the plug-and-play manager deriving the selected explanation module from among the explanation modules in the list, and the derived explanation module providing an explanation of the input artificial intelligence model to the user terminal.